curiositech/windags-skills

Bad-Faith Rhetoric Detector

When to Use This Skill

Load this skill when the user suspects manipulation, asks about "bad faith" responses, wants to analyze debate tactics, or uses phrases like "dodging the question," "moving goalposts," or "gaslighting."

DECISION POINTS

Primary Tactic Identification (when multiple patterns co-occur)

1. WHAT was the original question/claim?
   ↓
2. HOW did the response handle it?
   ├─ Completely ignored → Stonewalling or Deflection
   ├─ Replaced with different topic → Whataboutism or Gish Gallop
   ├─ Attacked the asking style → Tone Policing
   ├─ Retreated to safer claim → Motte-and-Bailey
   └─ Demanded excessive proof → Sealioning or Burden-Shifting
   ↓
3. WHAT was the emotional effect?
   ├─ Exhaustion from volume → Gish Gallop (primary)
   ├─ Self-doubt about reasonableness → DARVO or Gaslighting
   ├─ Confusion about what was agreed → Moving Goalposts
   └─ Frustration at non-engagement → Stonewalling (primary)

Confidence Assessment Tree

High Confidence (>80%):
├─ Pattern appears 3+ times in exchange
├─ Clear quotes demonstrate the behavior
└─ Effect on conversation is obvious

Medium Confidence (50-80%):
├─ Pattern appears 1-2 times
├─ Some ambiguity in interpretation
└─ Could be defensive behavior instead

Low Confidence (<50%):
├─ Single instance only
├─ Multiple reasonable interpretations
└─ Context suggests good faith possible

Response Strategy Selection

IF relationship is unsafe OR pattern is clearly intentional
   → Name + Exit (use boundary scripts)

ELSE IF pattern seems defensive/overwhelmed
   → Name + Pause ("Let's take a step back...")

ELSE IF user needs ongoing relationship
   → Name + Limit ("I'll engage with X, but not Y")

ELSE IF one-off interaction
   → Exit without explanation

FAILURE MODES

Anti-Pattern: "Pattern Confirmation Bias"

Symptom: Finding bad faith everywhere, even in ambiguous exchanges Detection Rule: If >70% of analyzed exchanges show "bad faith," you're overcalling Fix: Force charitable interpretation first, then pattern match second

Anti-Pattern: "Therapeutic Overreach"

Symptom: Diagnosing psychological conditions from rhetoric patterns Detection Rule: Using words like "narcissist," "sociopath," or "personality disorder" Fix: Stick to behavioral descriptions only: "This response pattern avoids..."

Anti-Pattern: "False Equivalence Detection"

Symptom: Labeling legitimate disagreement as "bad faith" Detection Rule: No actual question dodging, just unwelcome answer Fix: Distinguish between "didn't answer how I wanted" vs "didn't answer at all"

Anti-Pattern: "Weaponized Analysis"

Symptom: User wants help deploying these tactics themselves Detection Rule: Questions focus on "how to" rather than "is this happening" Fix: Decline to assist with tactical deployment; redirect to productive discourse

Anti-Pattern: "Analysis Paralysis Loop"

Symptom: Endless cataloging without decision or action Detection Rule: 3+ rounds of pattern identification with no boundary setting Fix: "What do you want to do with this information?" Force decision point

WORKED EXAMPLES

Example 1: Workplace Motte-and-Bailey

User: "My boss said in the meeting that our department 'needs major restructuring and some people aren't pulling their weight.' When I asked privately who specifically and what changes, he said 'I'm just saying we all need to step up our game.' Is this bad faith?"

Analysis Process:

1. Original claim: Department needs major restructuring + some people underperforming
2. Response to clarification request: Vague "we all need to step up"
3. Pattern match: Motte-and-Bailey (retreat from specific threat to generic encouragement)
4. Confidence: High (clear retreat pattern)
5. Alternative explanation: Boss may lack specific plan, using vague language to avoid premature commitments

Output: "This follows a motte-and-bailey pattern - making a specific concerning claim (major restructuring, people not pulling weight) then retreating to an uncontroversial position (everyone should work hard) when pressed for details. This could indicate he's testing reactions before committing to specifics, or he may lack a concrete plan but wants to create urgency."

Example 2: Relationship DARVO

User: "I told my partner I felt hurt when they canceled our plans last minute again. They said 'I can't believe you're attacking me when I'm dealing with work stress. You never support me when I'm struggling.' Did I do something wrong?"

Analysis Process:

1. Original concern: Pattern of last-minute plan cancellation causing hurt
2. Response pattern: Deny wrongdoing, attack user for raising it, position self as victim
3. Pattern match: DARVO (Deny-Attack-Reverse Victim/Offender)
4. Alternative explanation: Partner genuinely overwhelmed, defensive reaction to feeling criticized
5. Effect: User now doubting if their concern was legitimate

Output: "This response follows DARVO - your concern about canceled plans was denied ('can't believe you're attacking'), you were attacked for raising it, and they repositioned themselves as the victim who needs support. This could be manipulative, or it could be a defensive reaction from someone who feels overwhelmed and criticized. The effect either way is that your original concern (cancellation pattern) remains unaddressed."

QUALITY GATES

Analysis is complete when you can answer YES to all:

• Identified the original question/claim that was allegedly avoided or deflected
• Mapped how the response handled (or failed to handle) that original point
• Named specific pattern(s) with confidence level (high/medium/low)
• Provided at least one charitable alternative explanation
• Described observable effect on the conversation or relationship
• Checked for pattern confirmation bias (am I overcalling this?)
• Avoided psychological diagnosis (stuck to behavioral description)
• Offered practical next-step options for the user
• Verified quotes/examples support the pattern identification
• Considered context that might explain the behavior differently

NOT-FOR BOUNDARIES

Do NOT use this skill for:

• Helping users deploy bad-faith tactics themselves → Use productive-discourse-facilitator for ethical persuasion
• Diagnosing personality disorders → Refer to mental health professionals
• Analyzing your own responses in conversation → Use communication-clarity-optimizer for self-improvement
• Legal or HR advice about hostile communication → Refer to legal professionals
• Determining if someone is "lying" definitively → Use factual-accuracy-verifier for truth claims

Delegate when:

• User needs specific response scripts → Use boundary-setting-navigator
• Focus is on repairing damaged communication → Use relationship-repair-facilitator
• Analysis reveals potential abuse → Prioritize safety resources over pattern analysis